Frames vs. cells: Who cares? Part 1

A couple of weeks ago, we began a discussion about frames and cells. In that newsletter, we promised to look and see if the age-old arguments of whether frames or cells are “better” for your network really mattered any more.

A couple of weeks ago, we began a discussion about frames and cells.  In that newsletter, we promised to look and see if the age-old arguments of whether frames or cells are “better” for your network really mattered any more.

Of course, the fundamental perceived advantage of frames is that variable-length packets use WAN bandwidth more efficiently.  Transporting information in cells requires multiple cells where a single frame would do for larger chunks of information.  And for shorter chunks of information, the cells sometimes must be padded with electronic plastic peanuts to fill out the required cell length.

The bandwidth efficiency portion of this equation seems to have fallen by the wayside for the most part.  During the controversy’s heyday, branch offices were typically connected to the data center with 56K bit/sec circuits.  Today, thanks to technologies like DSL and cable, the typical connection speed is experiencing at least a 10-fold increase.  Also, the cost of transmission per bit per second continues to fall.

Perhaps the most telling empirical evidence that the overhead is not a significant factor comes from the popularity of TCP/IP.  If you’re simply comparing the bandwidth efficiency of frame relay vs. ATM, the frame relay’s frame format is the clear winner.  However, TCP/IP, while being a frame format, has quite heavy overhead compared with either frame relay or ATM.  And this overhead is especially heavy for real-time traffic that has short payloads – such as VoIP.

So the bottom line here seems to be that enterprise users have decided that transmission overhead is not the most important factor in designing networks.  This makes the efficiency arguments for frame-oriented protocols and against cell-oriented protocols meaningless.

Next time, we’ll see whether the pro-cell arguments have been rendered equally ineffective.